Feature extraction for image recognition and computer vision

Feature extraction and classifier design are two main processing blocks in all pattern recognition and computer vision systems. For visual patterns, extracting robust and discriminative features from image is the most difficult yet the most critical step. Several typical and advanced approaches of feature extraction from image are explored, some of which are analyzed in depth. Various techniques of feature extraction from image are organized in four categories: human expert knowledge based methods, image local structure based approaches, image global structure based techniques and machine learning based statistical approaches. We will show examples of applying these feature extraction approaches to solve problems of the image based biometrics, including fingerprint verification/identification and face detection/recognition. These illustrative application examples unveil the ideas, principles and advancements of feature extraction techniques and demonstrate their effectiveness and limitations in solving real-world problems.